Don't report any errors in lower_intrinsics.
#115602
Conversation
…typecked before.
|
(rustbot has picked a reviewer for you, use r? to override) |
rustbot
added
S-waiting-on-review
|
Some changes occurred to MIR optimizations cc @rust-lang/wg-mir-opt |
| if self.tcx().is_intrinsic(def_id) { | ||
| match self.tcx().item_name(def_id) { | ||
| sym::simd_shuffle => { | ||
| if !matches!(args[2], Operand::Constant(_)) { |
There was a problem hiding this comment.
I believe there are a couple of cases where stdarch has a shuffle where the last arg is not an Operand::Constant, but is still constant after LLVM level constprop. I do want this to be fixed though. Handling this in cg_clif requires a bunch of non-trivial code: https://github.com/bjorn3/rustc_codegen_cranelift/blob/0559de65672243a97d6cd6e6ee6a8e8c291ef4ce/src/constant.rs#L476C1-L575 (called from https://github.com/bjorn3/rustc_codegen_cranelift/blob/0559de65672243a97d6cd6e6ee6a8e8c291ef4ce/src/intrinsics/simd.rs#L171-L172) AFAIK stdarch's test suite doesn't get run in rust's CI. Only in the CI of rust-lang/stdarch.
There was a problem hiding this comment.
I just moved the check. We were already checking this. I think the const propagated ones haven't been working for a long time
There was a problem hiding this comment.
Looks like it is indeed unreachable for simd_shuffle now, but for simd_insert and simd_extract it is still possible and necessary for stdarch.
There was a problem hiding this comment.
every simd_insert usage outside of ui/codegen tests seems to be using constants
for simd_extract, everything but some extract helpers for #60637 are using constants
There was a problem hiding this comment.
and even those helpers are only used in tests
There was a problem hiding this comment.
In stdarch there are a couple of IMM8 as u32 which is not a constant. It has to be const { IMM8 as u32 }.
|
@bors r+ |
bors
added
S-waiting-on-bors
…henkov Don't report any errors in `lower_intrinsics`. Intrinsics should have been type checked earlier. This is part of moving all mir-opt diagnostics early enough so that they are reliably emitted even in check builds: rust-lang#49292 (comment)
|
☀️ Test successful - checks-actions |
|
Finished benchmarking commit (c5775a7): comparison URL. Overall result: no relevant changes - no action needed@rustbot label: -perf-regression Instruction countThis benchmark run did not return any relevant results for this metric. Max RSS (memory usage)ResultsThis is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.
CyclesResultsThis is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.
Binary sizeThis benchmark run did not return any relevant results for this metric. Bootstrap: 628.665s -> 629.749s (0.17%) |
Intrinsics should have been type checked earlier.
This is part of moving all mir-opt diagnostics early enough so that they are reliably emitted even in check builds: #49292 (comment)